Mangrove forest tree species classification method and device based on key meteorological event constraint

Abstract

The invention discloses a mangrove forest tree species classification method and device based on key meteorological event constraints, and the method comprises the following steps: enabling the classification process to focus on the response of a mangrove forest to a specific meteorological change through constructing a key meteorological event window with a time constraint effect; a tree species classification data set is constructed by taking the window as a constraint, so that a training sample closely corresponds to a key ecological process, and the identification effectiveness of different tree species is improved from the source; a classification model is trained by using the data set, so that the model can learn the differential performance of the mangrove forest under different meteorological conditions, and the space-time adaptability and the classification precision are enhanced; the trained model is applied to the to-be-classified area, and the tree species distribution result consistent with the key meteorological event window is output, so that the classification not only reflects the current vegetation situation, but also reflects the response rule of the mangrove forest to the meteorological driving mechanism, and the adaptability of the method to the meteorological change and the time sequence interpretation capability of the result are improved.

Description

Technical Field

[0001] This invention belongs to the fields of remote sensing geographic information systems and computer science, and more specifically, relates to a method and device for classifying mangrove tree species based on key meteorological events. Background Technology

[0002] Mangroves are a core component of coastal ecosystems, renowned as "coastal guardians" for their outstanding functions in windbreak and sand fixation, biodiversity conservation, and ecological environment regulation. Therefore, accurately monitoring the distribution and growth status of mangrove species is crucial for promoting resource conservation and achieving sustainable management. While remote sensing technology provides large-scale, continuous observation capabilities, greatly advancing mangrove research, its application faces significant challenges in terms of accuracy. Specifically, different mangrove species exhibit complex differences in age, growth status, and spectral characteristics, with small inter-species variations. This inherent spectral confusion makes high-precision, refined species classification based solely on remote sensing data extremely difficult.

[0003] However, a deeper and more critical challenge lies in the close relationship between mangrove growth and meteorological factors. Mangrove growth is a long-term, dynamic ecological process significantly driven by key meteorological factors such as precipitation, temperature, and wind speed. Different meteorological events, whether daily seasonal variations or sudden extreme events, directly lead to significant changes in the spectral characteristics and growth status of the vegetation.

[0004] The invention patent with publication number CN112861837B proposes a method for intelligent extraction of mangrove ecological information based on UAVs. This method first collects visible light remote sensing data from UAVs and preprocesses the data, then uses a pixel-level species identification algorithm to identify mangrove species and calculate mangrove area and species diversity. Then, it generates DSM results through 3D reconstruction and combines the mangrove plant species identification results to calculate ecological indicators such as canopy diameter and tree height of mangrove plants. However, this method does not introduce a key meteorological event constraint mechanism to screen remote sensing time-series features, and cannot fully explore the response differences of mangrove tree species under different meteorological conditions. Therefore, when facing complex environmental changes or extreme climate disturbances, the classification accuracy and model robustness are difficult to guarantee. Summary of the Invention

[0005] To overcome the problems of existing mangrove species classification methods lacking screening of remote sensing temporal features and having poor adaptability to meteorological changes, this invention provides a mangrove species classification method and device based on key meteorological event constraints.

[0006] The primary objective of this invention is to solve the aforementioned technical problems. The technical solution of this invention is as follows: The first aspect of this invention provides a method for classifying mangrove tree species based on key meteorological events, comprising the following steps: Time-series analysis of key meteorological factors for mangrove growth was conducted to extract key meteorological event window constraints; A tree species classification dataset was constructed using key meteorological event windows as constraints. The tree species classification model is trained using a tree species classification dataset to obtain a pre-trained model; The pre-trained model is used to classify tree species in the region to be classified, and the tree species distribution results are obtained.

[0007] Furthermore, the key meteorological elements include one or more of the following: temperature, precipitation, wind field, and radiation conditions; the key meteorological events include one or more of the following: typhoon season, high precipitation period during the rainy season, monsoon transition period, and drought period; the time-series analysis of key meteorological elements for mangrove growth includes the following steps: Anomaly analysis is used to calculate the deviation between the actual values ​​and long-term averages of key meteorological elements, identify anomalous years and extreme events, and output a sequence of anomalous events. Perform trend analysis on the abnormal event sequence and plot the long-term evolution curves of meteorological elements; Periodic analysis was used to analyze the long-term evolution curve to obtain seasonal and interannual variation characteristics. The key meteorological event window for mangrove growth is located using the aforementioned change characteristics. The expression is as follows:

[0008] in, Indicates meteorological elements in time The observed values, The average value of meteorological elements over n years. Represents the intensity of anomalies; This refers to the trend slope obtained from time series fitting within the current analysis window for trend analysis. The standard deviation of the long-term trend slope. This represents the significant upward or downward trend of meteorological variables during the analysis window period; This refers to the periodic residual term obtained through periodic analysis. The long-term standard deviation of the periodic residual term. Represents the intensity of periodic anomalies; This is the set comprehensive disturbance threshold.

[0009] Furthermore, the anomaly analysis identifies abnormal meteorological events by deviating from the annual average by more than a preset threshold; the trend analysis uses the Mankendall test to detect the monotonic trend of meteorological variables; and the periodic analysis uses the fast Fourier transform to identify the seasonal and interannual variation patterns of meteorological elements.

[0010] Furthermore, a tree species classification dataset is constructed using key meteorological event windows as constraints, including the following steps: Constrained by key meteorological event windows, and using a combination of expert visual interpretation and historical confidence analysis, the spatial distribution of mangrove tree species was determined based on high-resolution historical remote sensing imagery. The spatial distribution results are rasterized to generate label raster and registered with remote sensing images. Initial mangrove samples are obtained using non-overlapping sampling. The initial sample is used to extract multidimensional features within the corresponding event window. The multidimensional features are then input into the Relief feature selection algorithm, which outputs feature weights. Redundant and noisy features are removed based on the feature weights to obtain high-confidence mangrove species samples. The features of the mangrove tree species samples under different event windows are sorted, normalized, and paired to output a tree species classification dataset.

[0011] Furthermore, the extracted multidimensional features include one or more of the following: spectral features and topographic parameters. The spectral features include the normalized vegetation index, and the topographic parameters include one or more of the following: slope, aspect, and relative elevation. The multidimensional features are input into the Relief feature selection algorithm to output feature weights, including the following steps: Initialize weights for feature j =1; From the dataset, m samples are randomly selected, where the feature vector of sample i is... For each sample, find its nearest neighbors in both the same and different classes. Calculate the features by comparing the feature differences between the sample and its nearest neighbors in both classes. Changes in the weights of the discriminative ability for classification The expression is as follows:

[0012] Where m is the number of randomly selected samples. For the value of feature j of sample i, Let j be the value of feature j of sample i in the nearest neighbor sample of the same class; Let j be the value of feature j of sample i in the nearest neighbor sample of the different class; Change in weight Add the weights to the original weights to obtain the weights of feature j; The above process is executed iteratively to obtain the feature weights for each feature.

[0013] Furthermore, the tree species classification model is an XGBoost model, and the tree species classification model is trained using a tree species classification dataset, including the following steps: The tree species classification dataset is divided into training and validation sets according to a preset ratio; The training set is input into the XGBoost model, and the key parameters are optimized by iterative parameter tuning strategy. The learning rate is set to a and the number of decision trees is set to b. The loss function is minimized by 5-fold cross-validation. The time series weights of historical product priors are introduced, and the high confidence samples of the last k years are given n times the weight. The preliminary model is then output. Input the validation set into the initial model and output the model accuracy evaluation index. If the evaluation index does not reach the preset threshold, the model parameters are fine-tuned based on the error distribution of the validation set and combined with grid search. Iterative verification is performed until the evaluation index converges, and the fine-tuned tree species classification model is output.

[0014] Furthermore, the pre-trained tree species classification model is used to classify the tree species in the region to be classified, including the following steps: Preprocess multi-temporal remote sensing images of the region to be classified to output a spectrally consistent image set; The spectrally consistent image set was time-series matched according to the key meteorological event windows, and multi-dimensional features were extracted to obtain a multi-dimensional feature set; The multi-dimensional feature set is standardized and time-series paired, and the resulting feature matrix is ​​input into a pre-trained tree classification model to output preliminary class probabilities. The preliminary category probabilities were verified and corrected using historical product prior probabilities and field survey data to obtain the tree species distribution results.

[0015] Furthermore, preprocessing of multi-temporal remote sensing images of the region to be classified includes one or more of the following: radiometric calibration, atmospheric correction, and cloud and shadow removal.

[0016] Furthermore, the preliminary category probabilities are validated and revised using historical product prior probabilities combined with field survey data, including the following steps: Preliminary category probabilities Prior probability of historical products proportionally Linear fusion combines the fusion results with growth suitability or anomalous response characteristics within key meteorological event windows. We perform weighted calculations to obtain the optimized class probabilities. The expression is as follows:

[0017] use The corresponding tree species distribution map is compared with the field survey data to calculate the accuracy index. The accuracy index includes one or more of the following: overall accuracy, Kappa coefficient, user accuracy calculated by tree species, and mapping accuracy. The ratio is dynamically adjusted according to the accuracy index. and Repeat the above probability fusion and evaluation steps until the accuracy index converges, and output the final tree species distribution result.

[0018] A second aspect of the present invention provides an electronic device, including a memory and a processor, wherein the memory includes a program for a mangrove species classification method based on key meteorological event constraints, and when the program for a mangrove species classification method based on key meteorological event constraints is executed by the processor, the program implements the steps of a mangrove species classification method based on key meteorological event constraints.

[0019] Compared with the prior art, the beneficial effects of the technical solution of the present invention are: This invention constructs a key meteorological event window with time constraints, focusing the classification process on the mangrove response to specific meteorological changes, significantly improving the discriminative ability in the time dimension. By constructing a tree species classification dataset with this window as a constraint, the training samples closely correspond to key ecological processes, improving the identification effectiveness of different tree species from the source. By training the classification model using the above dataset, the model can learn the differentiated performance of mangroves under different meteorological conditions, enhancing its spatiotemporal adaptability and classification accuracy. By applying the trained model to the area to be classified and outputting tree species distribution results consistent with the key meteorological event window, the classification not only reflects the current vegetation status but also embodies the response patterns of mangroves to meteorological driving mechanisms, improving the method's adaptability to meteorological changes and the temporal interpretability of the results. Attached Figure Description

[0020] To make the objectives and technical solutions of this invention clearer, the following drawings are provided and described: Figure 1 A flowchart of a mangrove species classification method based on key meteorological events is provided for an embodiment of the present invention; Figure 2 This is a schematic diagram of the window for extracting key meteorological events provided in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the spatial distribution of high-confidence mangrove tree species provided in an embodiment of the present invention. Figure 4 This is a schematic diagram of unmanned aerial vehicle (UAV) field data collection provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the mangrove tree species classification results provided in an embodiment of the present invention. Detailed Implementation

[0021] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0023] Example 1: This invention provides a method for classifying mangrove tree species based on key meteorological events, such as... Figure 1 The diagram shows a flowchart of a mangrove species classification method based on key meteorological events. The specific steps are as follows: S1: Conduct time-series analysis on key meteorological factors for mangrove growth and extract key meteorological event window constraints, such as... Figure 2 As shown, the key meteorological elements include one or more of the following: precipitation, average temperature, extreme wind speed and radiation conditions; the key meteorological events include one or more of the following: typhoon season, rainy season with high precipitation, monsoon transition period, and drought period.

[0024] Mangrove growth is highly sensitive to meteorological factors such as precipitation, temperature, and wind speed. Fluctuations in these factors can directly cause abnormal responses in remote sensing spectra, thus affecting the accuracy of subsequent tree species classification. Therefore, before conducting mangrove tree species classification, it is essential to perform a systematic time-series analysis of key meteorological factors to define precise time ranges for subsequent research. Traditional methods often rely on fixed thresholds for temperature or precipitation to directly define critical periods, but this approach has significant drawbacks: firstly, it easily leads to data redundancy, introducing a large amount of irrelevant interference information; secondly, it is difficult to avoid noise in meteorological data, failing to truly capture the "effective window" of mangrove response to meteorological factors. To address this issue, this invention employs a time-series method combining anomaly analysis, trend analysis, and periodic analysis to comprehensively analyze meteorological factors related to mangrove growth in the study area from multiple dimensions, ultimately extracting key meteorological factors closely related to mangrove growth (such as precipitation, average temperature, extreme wind speed, and radiation conditions) and their corresponding key time periods (such as typhoon peak periods, high precipitation periods during the rainy season, monsoon transition periods, and drought periods). The specific process is as follows: S1.1: Anomaly analysis is used to calculate the deviation between actual values ​​of meteorological elements and long-term averages, identify anomalous years and extreme events, and output anomaly event sequences. In this embodiment, based on observational data of precipitation, temperature, and wind speed from meteorological stations in Guangdong Province from 2018 to 2023, a systematic time series analysis is conducted using Python. Anomaly analysis is used to identify anomalous years and extreme events such as typhoons and droughts. Anomaly analysis refers to comparing observed values ​​(such as temperature and precipitation) of a specific period (such as a day, month, or year) with a long-term reference benchmark (usually the multi-year average, called the "climate average") to characterize the degree of deviation of meteorological elements from the multi-year average. Anomalous meteorological events are identified when the deviation from the multi-year average exceeds an anomaly threshold, reflecting the intensity of disturbance of meteorological factors on the mangrove growth environment.

[0025] S1.2: Perform trend analysis on the abnormal event sequence and plot the long-term evolution curves of meteorological elements. Use the long-term evolution curves of meteorological elements to characterize the annual average temperature change trend and the seasonal distribution of precipitation, as a regional climate background.

[0026] S1.3: Use periodic analysis to perform periodic analysis on long-term evolution curves to obtain seasonal (e.g., rainy / dry season precipitation fluctuations) and interannual variation characteristics (e.g., the periodic impact of El Niño on temperature). S1.4: Utilize the aforementioned change characteristics to locate the key meteorological event window for mangrove growth, and output the key meteorological event window constraints. The expression is as follows:

[0027] in, Indicates meteorological elements in time The observed values, The average value of meteorological elements over n years. It represents the intensity of the anomaly and is used to measure the relative deviation of meteorological data at that moment from the multi-year average. This refers to the trend slope obtained from time series fitting within the current analysis window for trend analysis. The standard deviation of the long-term trend slope. This represents the significant upward or downward trend of meteorological variables during the analysis window period; This refers to the periodic residual term obtained through periodic analysis. The long-term standard deviation of the periodic residual term. Represents the intensity of periodic anomalies; The threshold for comprehensive disturbance is set. In this embodiment, anomaly analysis is used to characterize the degree of deviation of meteorological elements from the multi-year average. An abnormal meteorological event is identified when the deviation from the 5-year average of 2018-2023 exceeds 10%, thereby reflecting the intensity of disturbance of meteorological factors on the mangrove growth environment. Trend analysis uses the Mann-Kendall test to detect the monotonic trend of meteorological variables, with a significance level of less than 0.05 as the criterion for determining the significance of the trend. Periodic analysis uses Fast Fourier Transform (FFT) to identify the seasonal and interannual variation patterns of meteorological elements.

[0028] In step S1 above, various meteorological response characteristic indicators are extracted from dimensions such as temperature, precipitation, wind field, and radiation conditions. These include temperature characteristics such as average temperature, maximum temperature, minimum temperature, temperature anomaly, and accumulated temperature; and precipitation characteristics such as total precipitation, standardized precipitation index (SPI), number of consecutive precipitation days, and extreme precipitation intensity. The temporal correspondence between mangrove growth and typical meteorological events in Guangdong (such as typhoon season and drought period) is clarified, and the key meteorological elements affecting mangrove growth and their periods of influence are determined. Based on the above analysis results, a spatiotemporal constraint basis is provided for subsequent multi-source feature construction and tree species classification, supporting the accurate remote sensing identification and classification of mangroves in Guangdong Province.

[0029] S2: Construct a tree species classification dataset with key meteorological event windows as constraints.

[0030] The specific process is as follows: S2.1: As Figure 3 As shown, constrained by critical meteorological event windows, this study utilizes a combination of expert visual interpretation and historical confidence analysis to determine the spatial distribution of high-confidence mangrove tree species based on high-resolution historical remote sensing imagery. In this embodiment, Sentinel-2 optical imagery, Sentinel-1 radar data, and elevation data acquired through the GEE platform are combined to perform refined visual interpretation of mangrove areas in Guangdong Province, marking the spatial distribution of different tree species within critical meteorological event windows. Remote sensing images within the region are significantly affected by meteorological disturbances, and can better reflect the spectral and structural response characteristics of mangroves under extreme precipitation, typhoon wind speeds, and high-temperature stress. This invention selects only those located within... The system interprets and constructs samples from multi-source image datasets, and achieves physical constraints between meteorological events and remote sensing observations through the spatiotemporal matching of meteorological windows and multi-source images.

[0031] S2.2: Rasterize the spatial distribution results to generate label raster and register them with remote sensing images. Use non-overlapping sampling on the GEE platform to obtain initial mangrove samples.

[0032] S2.3: Extract multidimensional features (including spectral features and terrain parameters, where the spectral features include the Normalized Difference Vegetation Index (NDVI) and the terrain parameters include slope, aspect, and relative elevation) from the initial samples within the corresponding event window. Input the multidimensional features into the Relief feature selection algorithm and output feature weights, which reflect the contribution of the feature in distinguishing different mangrove species. Based on the feature weights, remove redundant and noisy features and retain only high-weight features to construct high-confidence mangrove species samples. These feature weights are not only used for feature selection but also provide weighted inputs in the subsequent classifier.

[0033] More specifically, the method for inputting multidimensional features into the feature weights output by the Relief feature selection algorithm is as follows: S2.3.1: Initialize weights for feature j =1; S2.3.2: Randomly select m samples from the dataset, where the feature vector of sample i is... For each sample, find its nearest neighbors in both the same and different classes. Calculate the features by comparing the feature differences between the sample and its nearest neighbors in both classes. Changes in the weights of the discriminative ability for classification The expression is as follows:

[0034] Where m is the number of randomly selected samples. For the value of feature j of sample i, Let j be the value of feature j of sample i in the nearest neighbor sample of the same class; Let j be the value of feature j of sample i in the nearest neighbor sample of the different class; S2.3.3: Change in weight Add the weights to the original weights to obtain the weights of feature j; S2.3.4: Iteratively execute the above process to obtain the feature weight of each feature.

[0035] S2.4: Organize and normalize the features of the mangrove tree species samples under different event windows to eliminate differences in dimensionality, and pair the features under each window to ensure their consistency and comparability in the temporal dimension, outputting a tree species classification dataset with complete structure and standardization, providing a reliable data foundation for subsequent model training.

[0036] Step S2 above not only ensures that the classification model can fully learn the differentiated response characteristics of mangrove tree species under different meteorological conditions, but also effectively improves the spatiotemporal coverage and representativeness of multi-source features, providing reliable and solid data support for subsequent tree species classification.

[0037] S3: Train the tree species classification model using the tree species classification dataset to obtain a pre-trained model.

[0038] More specifically, the tree species classification model is an XGBoost model, and the tree species classification model is trained using a tree species classification dataset, including the following steps: S3.1: Divide the tree species classification dataset into training and validation sets according to a preset ratio (7:3 in this embodiment), and ensure that the samples are evenly distributed in different climate zones and seasons to provide high-quality input for XGBoost model training; S3.2: Receive data from different key weather event windows The multi-source features are used as model input. For each sample... The input vector is:

[0039] in, This represents the feature set of the sample within the k-th meteorological event window, including spectral, topographic, and time-series response features. The multi-source features of the training set are input into the XGBoost model. During the model training phase, an iterative parameter tuning strategy is used to optimize key parameters: the learning rate is set to 0.05 to balance training efficiency and accuracy, the number of decision trees is set to 300 to avoid overfitting, and the loss function is minimized through 5-fold cross-validation. Simultaneously, prior time-series weights from historical products are introduced, assigning n (user-defined, 2 in this example) times the weight to high-confidence samples from the last k years (5 in this example), making the model more focused on the growth and distribution patterns of mangroves under recent climatic conditions, and outputting a preliminary model. S3.3: Input the validation set into the preliminary model and output the model accuracy evaluation index (overall accuracy and Kappa coefficient). If the evaluation index does not reach the preset threshold, the model parameters are fine-tuned based on the error distribution of the validation set and combined with grid search. Iterative verification is performed until the evaluation index converges. The maximum tree depth (max_depth) is determined to be 6 and the minimum number of child splits (min_child_weight) is determined to be 3. The fine-tuned XGBoost mangrove tree species classification model is output. The pre-trained mangrove tree species classification model obtained through this process can deeply integrate time series information (changes in vegetation index in different growing seasons) and spatial distribution features (spatial correlation of topography and hydrology). It not only achieves high-precision identification, but also significantly improves the model's stability and cross-regional generalization ability under extreme climate and complex terrain through the dual guarantee of meteorological constraints and historical priors, laying a technical foundation for subsequent large-scale mangrove monitoring.

[0040] S4: Use the pre-trained model to classify tree species in the region to be classified and obtain the tree species distribution results.

[0041] The specific process is as follows: S4.1: Multi-temporal remote sensing image data of the target area in Guangdong Province to be classified are preprocessed on the GEE platform to output a spectrally consistent image set.

[0042] More specifically, the preprocessing method includes one or more of the following: radiometric calibration, atmospheric correction, and cloud and shadow removal.

[0043] S4.2: Perform time-series matching of the spectrally consistent image set according to the key meteorological event window to make its distribution consistent with the training sample, and extract multi-dimensional features to obtain a multi-dimensional feature set.

[0044] S4.3: Standardize and temporally pair the multi-dimensional feature set, input the resulting feature matrix into the pre-trained XGBoost model, output preliminary class probabilities, and use them for subsequent probability fusion.

[0045] S4.4: Utilize prior probabilities from historical products combined with field survey data to verify and correct preliminary category probabilities, output tree species distribution results, ensuring that the tree species distribution results are spatially continuous and complete, truly reflecting the differentiated responses of different mangrove tree species in Guangdong Province under the constraints of key meteorological events, and providing high-precision basic data for regional mangrove monitoring and ecological management.

[0046] The specific steps are as follows: Preliminary category probabilities Prior probability of historical products proportionally Linear fusion combines the fusion results with growth suitability or anomalous response characteristics within key meteorological event windows. We perform weighted averages to obtain the optimized class probabilities. The expression is as follows:

[0047] Where i is the spatial pixel number and c is the mangrove tree species number. In this way, the classification results can more accurately reflect the actual distribution of mangroves under extreme weather conditions while maintaining spatial continuity.

[0048] use Corresponding tree species distribution map and field survey data (using drones to collect field data to determine high-confidence mangrove areas and the distribution of various tree species, such as...) Figure 4 The accuracy indicators are compared with those shown below, and the accuracy indicators include one or more of the following: overall accuracy, Kappa coefficient, user accuracy calculated by tree species, and mapping accuracy. The ratio is dynamically adjusted according to the accuracy index. and Repeat the above probability fusion and evaluation steps until the accuracy index converges, and output the verified and corrected spatial distribution results of mangrove tree species, such as... Figure 5 The diagram shows the results of mangrove species classification, using Kandelia candel and Avicennia marina as examples.

[0049] The mangrove species classification and mapping results obtained in step S4 above not only have continuity and completeness in spatial distribution, but also truly reflect the differentiated response patterns of different tree species under the constraints of key meteorological events, providing high-precision basic data support for mangrove monitoring and ecological management at regional and even global scales.

[0050] Experimental results in Guangdong Province demonstrate that the method of this invention exhibits significant advantages over traditional methods in mangrove species classification: In terms of classification accuracy, the overall accuracy reaches 99.2%, with a Kappa coefficient of 0.93, an improvement of approximately 5–8 percentage points compared to traditional methods; regarding ecological rationality, the optimized classification results highly match the actual distribution patterns of mangroves, such as the main distribution of *Kandelia candel* in the lower intertidal zone and *Avicennia marina* concentrated in the high tide zone, reflecting the ecological consistency of the classification results; in terms of meteorological adaptability, the event window mechanism effectively characterizes the temporal variation of species characteristics under different meteorological conditions, especially after typhoon events, the classification accuracy of *Rhizophora stylosa* is significantly improved. These results fully verify the practicality and advancement of the method of this invention.

[0051] Example 2: This embodiment provides an electronic device, including a memory and a processor. The memory includes a program for a mangrove species classification method based on key meteorological events constraints. When the processor executes the program for the mangrove species classification method based on key meteorological events constraints, it implements the steps of a mangrove species classification method based on key meteorological events constraints as described in Embodiment 1.

[0052] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively describe all embodiments here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.

Claims

1. A method for classifying mangrove tree species based on key meteorological events, characterized in that, Includes the following steps: Time-series analysis of key meteorological elements for mangrove growth was conducted to extract key meteorological event windows; A tree species classification dataset was constructed using key meteorological event windows as constraints. The tree species classification model is trained using a tree species classification dataset to obtain a pre-trained model; The pre-trained model is used to classify tree species in the region to be classified, and the tree species distribution results are obtained.

2. The mangrove species classification method based on key meteorological events as described in claim 1, characterized in that, The key meteorological elements include one or more of the following: temperature, precipitation, wind field, and radiation conditions; the key meteorological events include one or more of the following: typhoon season, high precipitation period during the rainy season, monsoon transition period, and drought period; the time-series analysis of the key meteorological elements for mangrove growth includes the following steps: Anomaly analysis is used to calculate the deviation between the actual values ​​and long-term averages of key meteorological elements, identify anomalous years and extreme events, and output a sequence of anomalous events. Perform trend analysis on the abnormal event sequence and plot the long-term evolution curves of meteorological elements; Periodic analysis was used to analyze the long-term evolution curve to obtain seasonal and interannual variation characteristics. The key meteorological event window for mangrove growth is located using the aforementioned change characteristics. The expression is as follows: in, Indicates meteorological elements in time The observed values, The average value of meteorological elements over n years. Represents the intensity of anomalies; This refers to the trend slope obtained from time series fitting within the current analysis window for trend analysis. The standard deviation of the long-term trend slope. This represents the significant upward or downward trend of meteorological variables during the analysis window period; This refers to the periodic residual term obtained through periodic analysis. The long-term standard deviation of the periodic residual term. Represents the intensity of periodic anomalies; This is the set comprehensive disturbance threshold.

3. The mangrove species classification method based on key meteorological event constraints according to claim 2, characterized in that, The anomaly analysis identifies abnormal meteorological events by deviating from the annual average value by more than a preset threshold. The trend analysis uses the Mankendall test to detect the monotonic trend of meteorological variables; the periodic analysis uses the fast Fourier transform to identify the seasonal and interannual variation patterns of meteorological elements.

4. The mangrove species classification method based on key meteorological event constraints according to claim 1, characterized in that, The tree species classification dataset is constructed using key meteorological event windows as constraints, including the following steps: Constrained by key meteorological event windows, and using a combination of expert visual interpretation and historical confidence analysis, the spatial distribution of mangrove tree species was determined based on high-resolution historical remote sensing imagery. The spatial distribution results are rasterized to generate label raster and registered with remote sensing images. Initial mangrove samples are obtained using non-overlapping sampling. The initial sample is used to extract multidimensional features within the corresponding event window. The multidimensional features are then input into the Relief feature selection algorithm, which outputs feature weights. Redundant and noisy features are removed based on the feature weights to obtain high-confidence mangrove species samples. The features of the mangrove tree species samples under different event windows are sorted, normalized, and paired to output a tree species classification dataset.

5. The mangrove species classification method based on key meteorological events as described in claim 4, characterized in that, The extracted multidimensional features include one or more of the following: spectral features and topographic parameters. The spectral features include the normalized vegetation index, and the topographic parameters include one or more of the following: slope, aspect, and relative elevation. The multidimensional features are input into the Relief feature selection algorithm to output feature weights. Includes the following steps: Initialize weights for feature j =1; From the dataset, m samples are randomly selected, where the feature vector of sample i is... For each sample, find its nearest neighbors in both the same and different classes. Calculate the features by comparing the feature differences between the sample and its nearest neighbors in both classes. Changes in the weights of the discriminative ability for classification The expression is as follows: Where m is the number of randomly selected samples. For the value of feature j of sample i, Let j be the value of feature j of sample i in the nearest neighbor sample of the same class; Let j be the value of feature j of sample i in the nearest neighbor sample of the different class; Change in weight Add the weights to the original weights to obtain the weights of feature j; The above process is executed iteratively to obtain the feature weights for each feature.

6. The mangrove species classification method based on key meteorological event constraints according to claim 1, characterized in that, The tree species classification model is an XGBoost model. The model is trained using a tree species classification dataset, and includes the following steps: The tree species classification dataset is divided into training and validation sets according to a preset ratio; The training set is input into the XGBoost model, and the key parameters are optimized by iterative parameter tuning strategy. The learning rate is set to a and the number of decision trees is set to b. The loss function is minimized by 5-fold cross-validation. The time series weights of historical product priors are introduced, and the high confidence samples of the last k years are given n times the weight. The preliminary model is then output. Input the validation set into the initial model and output the model accuracy evaluation index. If the evaluation index does not reach the preset threshold, the model parameters are fine-tuned based on the error distribution of the validation set and combined with grid search. Iterative verification is performed until the evaluation index converges, and the fine-tuned tree species classification model is output.

7. The mangrove species classification method based on key meteorological event constraints according to claim 1, characterized in that, The pre-trained tree species classification model is used to classify tree species in the region to be classified, including the following steps: Preprocess multi-temporal remote sensing images of the region to be classified to output a spectrally consistent image set; The spectrally consistent image set was time-series matched according to the key meteorological event windows, and multi-dimensional features were extracted to obtain a multi-dimensional feature set; The multi-dimensional feature set is standardized and time-series paired, and the resulting feature matrix is ​​input into a pre-trained tree classification model to output preliminary class probabilities. The preliminary category probabilities were verified and corrected using historical product prior probabilities and field survey data to obtain the tree species distribution results.

8. The mangrove species classification method based on key meteorological event constraints according to claim 7, characterized in that, Preprocessing of multi-temporal remote sensing images of the region to be classified includes one or more of the following: radiometric calibration, atmospheric correction, and cloud and shadow removal.

9. The mangrove species classification method based on key meteorological event constraints according to claim 7, characterized in that, The preliminary category probabilities are validated and revised using historical product prior probabilities combined with field survey data, including the following steps: Preliminary category probabilities Prior probability of historical products proportionally Linear fusion combines the fusion results with growth suitability or anomalous response characteristics within key meteorological event windows. We perform weighted summation to obtain the optimized class probabilities. The expression is as follows: use The corresponding tree species distribution map is compared with the field survey data to calculate the accuracy index. The accuracy index includes one or more of the following: overall accuracy, Kappa coefficient, user accuracy calculated by tree species, and mapping accuracy. The ratio is dynamically adjusted according to the accuracy index. and Repeat the above probability fusion and evaluation steps until the accuracy index converges, and output the final tree species distribution result.

10. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory includes a program for a mangrove species classification method based on key meteorological events. When the program for a mangrove species classification method based on key meteorological events is executed by the processor, it implements the steps of a mangrove species classification method based on key meteorological events as described in any one of claims 1 to 9.

Images